Ref T2015. This allows for Drydock blueprints to lease and acquire resources during their own leasing or resource acquisition. This is reqiured for working copies, which must lease a host resource.
Details
Details
- Reviewers
epriestley - Group Reviewers
Blessed Reviewers - Maniphest Tasks
- T2015: Implement Drydock
Tested as part of the working copy changes.
Diff Detail
Diff Detail
- Repository
- rP Phabricator
- Branch
- hachque-reconstructed
- Lint
Lint Passed - Unit
Tests Passed - Build Status
Buildable 7781 Build 8594: [Placeholder Plan] Wait for 30 Seconds Build 8593: arc lint + arc unit
Event Timeline
Comment Actions
I think this is obsoleted by the general de-nesting of allocation and acquisition and the clarification of locking/constraint responsibility.
Specifically, it is now expected that the allocator may build a resource for lease A and have it stolen by lease B before it can acquire a lease on it (which is fine, it will just retry a little later). And it is no longer expected that a single process can push an allocation through by performing less work than a generic taskmaster -- that is, we don't know which tasks need to be completed in order to complete an allocation or acquisition, only that doing work from the queue will eventually result in success if there are no actual configuration/implementation problems.
HEAD also now has a WorkingCopy blueprint which appears to work, lease host resources correctly, and not have race/constraint issues.
Let me know if I missed something and this is still relevant?